Message-ID: <CAKfTPtAXqVZvWynxazT+PYWptXm-ud7m64n2Zvd0XO4BrceMyQ@mail.gmail.com>
Date:   Wed, 1 Dec 2021 17:19:52 +0100
From:   Vincent Guittot <vincent.guittot@...aro.org>
To:     Vincent Donnefort <Vincent.Donnefort@....com>
Cc:     Valentin Schneider <Valentin.Schneider@....com>,
        peterz@...radead.org, mingo@...hat.com,
        linux-kernel@...r.kernel.org, mgorman@...hsingularity.net,
        dietmar.eggemann@....com
Subject: Re: [PATCH] sched/fair: Fix detection of per-CPU kthreads waking a task

On Wed, 1 Dec 2021 at 15:40, Vincent Donnefort
<vincent.donnefort@....com> wrote:
>
> On Tue, Nov 30, 2021 at 04:42:03PM +0100, Vincent Guittot wrote:
> > On Mon, 29 Nov 2021 at 17:54, Vincent Donnefort
> > <vincent.donnefort@....com> wrote:
> > >
> > > [...]
> > >
> > > > > > >
> > > > > > > still i don't see the need of !is_idle_task(current)
> > > > > > >
> > > > > >
> > > > > > Admittedly, belts and braces. The existing condition checks rq->nr_running <= 1
> > > > > > which can lead to coscheduling when the wakeup is issued by the idle task
> > > > > > (or even if rq->nr_running == 0, you can have rq->ttwu_pending without
> > > > > > having sent an IPI due to polling). Essentially this overrides the first
> > > > > > check in sis() that uses idle_cpu(target) (prev == smp_processor_id() ==
> > > > > > target).
> > > > > >
> > > > > > I couldn't prove such wakeups can happen right now, but if/when they do
> > > > > > (AIUI it would just take someone to add a wake_up_process() down some
> > > > > > smp_call_function() callback) then we'll need the above. If you're still
> > > > > > not convinced by now, I won't push it further.
> > > > >
> > > > > From a quick experiment, even with the asym_fits_capacity(), I can trigger
> > > > > the following:
> > > > >
> > > > > [    0.118855] select_idle_sibling: wakee=kthreadd:2 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > > > [    0.128214] select_idle_sibling: wakee=rcu_gp:3 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > > > [    0.137327] select_idle_sibling: wakee=rcu_par_gp:4 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > > > [    0.147221] select_idle_sibling: wakee=kworker/u16:0:7 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > > > [    0.156994] select_idle_sibling: wakee=mm_percpu_wq:8 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > >
> > > > Timestamp shows its booting phase and thread name above shows per cpu
> > > > thread. Could it happen just while creating per cpu thread at boot and
> > > > as a result not relevant ?
> > >
> > > I have more of those logs a bit later in the boot:
> > >
> > > [    0.484791] select_idle_sibling: wakee=kthreadd:2 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > [    0.516495] select_idle_sibling: wakee=kthreadd:2 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > [    0.525758] select_idle_sibling: wakee=kthreadd:2 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > [    0.535078] select_idle_sibling: wakee=kthreadd:2 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > [    0.547486] select_idle_sibling: wakee=kthreadd:2 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > > [    0.579192] select_idle_sibling: wakee=kthreadd:2 nr_cpus_allowed=8 current=swapper/0:1 in_task=1
> > >
> > > The nr_cpus_allowed=8 suggest that none of the threads from the logs I
> > > shared are per-CPU. Sorry if the format is confusing, I used:
> > >
> > >   wakee=<comm>:<pid> current=<comm>:<pid>.
> > >
> > > >
> > > > Can you see similar things later after booting ?
> > >
> > > I tried few scenarios other than the boot time but none of them produced
> > > "current=swapper/X:1 in_task=1"
> > >
> > > >
> > > > I have tried to trigger the situation but failed to get wrong
> > > > sequence. All are coming from interrupt while idle.
> > > > After adding in_task() condition, I haven't been able to trigger the
> > > > warn() that I added to catch the wrong situations on SMP, Heterogenous
> > > > or NUMA system. Could you share more details on your setup ?
> > > >
> > >
> > > This is just my Hikey960 with the asym_fits_capacity() fix [1] to make sure I
> > > don't simply hit the other issue with asym platforms.
> >
> > I ran my previous tests on dragonboard 845c which is dynamiQ and I
> > have tried on my hikey960 since  but without any success so far. This
> > is what i use:
> >
> > --- a/kernel/sched/fair.c
> > +++ b/kernel/sched/fair.c
> > @@ -6397,9 +6397,12 @@ static int select_idle_sibling(struct
> > task_struct *p, int prev, int target)
> >          * essentially a sync wakeup. An obvious example of this
> >          * pattern is IO completions.
> >          */
> > -       if (is_per_cpu_kthread(current) &&
> > +       if (in_task() &&
> > +           is_per_cpu_kthread(current) &&
> >             prev == smp_processor_id() &&
> >             this_rq()->nr_running <= 1) {
> > +
> > +               WARN(is_idle_task(current), "idle per cpu kthread: cpu
> > %d task: %s", prev, p->comm);
> >                 return prev;
> >         }
> >
> >
> > Without in_task() condition, i've got warnings from interrupt context
> > but nothing else.
> > Note that I don't even have the asym_fits_capacity() condition
>
> I could not find a setup reproducing that issue outside of the boot time. So
> following our conversation, I made a v2 that switch !is_idle_task() to in_task().

Ok.
Thanks

>
> >
> > >
> > > Then I just added my log in the per-CPU kthread wakee stacking exit path
> > >
> > >     printk("%s: wakee=%s:%d nr_cpus_allowed=%d current=%s:%d in_task=%d\n",
> > >             __func__, p->comm, p->pid, p->nr_cpus_allowed, current->comm, current->pid, in_task());
> > >
> > >
> > > [1] https://lore.kernel.org/all/20211125101239.3248857-1-vincent.donnefort@arm.com/
> > >
> > >
> > > From the same logs I also see:
> > >
> > >   wakee=xfsaild/mmcblk0:4855 nr_cpus_allowed=8 current=kworker/1:1:1070 in_task=0
> > >
> > > Doesn't that look like a genuine wakeup that would escape the per-CPU kthread
> > > stacking exit path because of the in_task test?
>
> My bad, I checked and this is not a genuine one...
>

Powered by blists - more mailing lists